Content
52%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The router core is genuinely strong — a clear priority ladder, hard validation gates, and executable commands — but the body carries substantial repetition and retired-feature history that bloat token cost, and its progressive-disclosure structure is broken: every referenced path points outside the bundle while the bundle's real reference files are orphaned. Linking the existing references/ files and moving the harness/todo detail out of SKILL.md would fix both weaknesses.
Suggestions
Link the existing bundle references from the body (e.g. mx tag routing → references/mx-tag.md, review/run quality guidance → references/anti-patterns.md, execution patterns → references/reference.md) so they are discoverable, and reconcile or remove the dangling workflows/*.md and .moai/.claude paths that are absent from the bundle.
Move the harness Branch A/A.1/B detail and the todo database-path policy into separate workflow/reference files, keeping only the argument-branching routing rule inline.
Trim historical and duplicated content — the retired --worktree/MODE_TEAM_UNAVAILABLE history, the repeated '(experimental — Agent Teams re-allowed; see Execution Mode Flags)' notes, and the duplicated routing-ledger policy — into a single 'Deprecated / History' section or delete it, and define the undefined placeholders (--tier, --level, --session).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~400-line body is noticeably verbose: it repeats the routing-ledger privacy policy in two places, restates '(experimental — Agent Teams re-allowed; see Execution Mode Flags)' on at least six workflow entries, and carries historical sediment ('the retired era emitted MODE_TEAM_UNAVAILABLE ... the sentinel is retained as documented history', the retired --worktree era) plus inline Korean error-message templates that belong in a reference file. It avoids a 1 because it never explains concepts Claude already knows — everything is project-specific policy — but much of the rationale and history could be trimmed or moved to a deprecated/legacy section. | 2 / 5 |
Actionability | The body provides copy-paste-ready commands ('echo "<raw request text>" | moai harness ledger record --subcommand <matched> ...', 'moai paths', 'moai todo list --json'), an explicit forbidden flag/subcommand table, exact error-message templates, and a concrete 10-step execution directive. Not a 5 because several placeholders ('--tier <tier>', '--level <harness-level>', '--session <session-id>') are never defined, and the Socratic/AskUserQuestion guidance stays high-level in places. | 4 / 5 |
Workflow Clarity | Routing follows a strictly ordered priority ladder (Priority 1 first-word matching → SPEC-ID detection → semantic classification → AskUserQuestion fallback) with explicit validation (Step 1.5 stops all processing on a forbidden flag/subcommand combination) and error-recovery templates, plus [HARD] markers acting as checkpoints. It falls short of 5 due to minor gaps: no check that 'workflows/<name>.md' exists before reading, and validation for destructive delegated workflows (harness remove, dead-code clean) is only implied via the downstream workflow files. | 4 / 5 |
Progressive Disclosure | The body's navigation is disconnected from the actual bundle: it directs the agent to 'Read workflows/plan.md' (and 17 other workflows/*.md paths) and to .moai/config and .claude/rules files, none of which exist in the bundle, while the four files that DO exist in references/ (reference.md, anti-patterns.md, mx-tag.md, file-reading-optimization.md) are never mentioned or linked from SKILL.md and are therefore undiscoverable. Large inline blocks (the three harness branches, the todo database-path policy) also clearly belong in separate files, matching anchor 2's 'content that clearly belongs in separate files is inlined; or references are buried' more than anchor 3's merely-imperfect organization. | 2 / 5 |
Total | 12 / 20 Passed |